We evaluate seven popular group fairness measures for student performance prediction problems. We conduct experiments using four traditional ML models and two fairness-aware ML methods on five educational datasets. Besides, we investigate the effect of varying grade thresholds on the accuracy and fairness of ML models. The preliminary results suggest that choosing the threshold is an important factor contributing to ensuring fairness in the output of the ML models.

Tai Le Quy, Thi Huyen Nguyen, Gunnar Friege and Eirini Ntoutsi “Evaluation of group fairness measures in student performance prediction problems”. 2022. In: Proceedings of SoGood 2022 – 7th Workshop on Data Science for Social Good Affiliated with ECML-PKDD 2022. Grenoble, France